Intro To Deep Learning

Leon Noel

RC Ascend

Agenda

  • Questions? 

  • Let's Talk -  Learning

  • Learn - What is Machine, Deep, Ect Learning?

  • Learn - What is a model?

  • Learn - Bird or Not Model?

  • Do - Read Paper + Watch 1.2

  • Homework - Week 01 Fast.ai

Questions

About last class or life

Resetting Forgetting Curve

https://www.pearsoned.com/three-simple-research-based-ways-to-ace-a-test

Active Recall & Spaced Repetition

Ali Abdaal: https://youtu.be/Z-zNHHpXoMM

Trough Of Sorrow

Manage Frustration

Consistency 

Taking Care Of Yourself

 

AI

A computer will do what you tell it to do.

What is artificial intelligence?

 Creation of computer systems that can perform tasks typically requiring human intelligence

Classification

Prediction

DATA

Prediction

What is machine learning?

 Creation of computer systems that can perform a task without be explicitly programmed to complete that task that typically requires structured data

https://www.kaggle.com/datasets/alessiocorrado99/animals10

cat

cat

cat

elephant

elephant

elephant

What is deep learning?

 Creation of computer systems that can perform a task without be explicitly programmed to complete that task that typically DOES NOT require 

structured data

https://www.kaggle.com/datasets/alessiocorrado99/animals10

DATA

Prediction

Training Set

Test Set

DATA

Training

Data

Test

Data

78 %

Algorithm

Algorithm

Linear Regression

Decision Tree

Manual Feature Engineering

Algorithm

Convolutional Neural Networks (CNNs)

Automatic Feature Engineering

Training

Data

Test

Data

78 %

Algorithm

Error Function

accuracy_score(prediction, test_data)

Output = Model

Model by itself is just an empty brain that has not yet learned anything

Weights represent the knowledge that the model has learned from the

training data

The primary function of a model is to make predictions or classifications based on input data. The weights are what allow the model to perform this function accurately. Without weights, the model cannot process the input data correctly to produce meaningful outputs.

Model trained to recognize cars will have weights that enable it to distinguish different cars

Model + Weights = Accurate Predictions

hopefully

Fine-tuning?

adjusting weights to minimize error

Bird or Not

What would be the Machine Learning approach?

What would be the Deep Learning approach?

Training

vs

Inference

Training

Training is the process of teaching the model using historical data, adjusting weights to minimize error

What would training look like for

"Bird or Not"

Inference

Inference is using the trained model to make predictions on new data.

What is used in production?

Homework

 

Watch: Fast.ai 1.2

Read: Overton 2019 - https://arxiv.org/abs/1909.05372